Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Do the reverse embedding in the same dtype as the input embedding #1548

Merged
merged 1 commit into from
Apr 10, 2024
Merged

Do the reverse embedding in the same dtype as the input embedding #1548

merged 1 commit into from
Apr 10, 2024

Conversation

mattdangerw
Copy link
Member

@mattdangerw mattdangerw commented Apr 3, 2024 •
edited
Loading

Fixes #1542

tirthasheshpatel
tirthasheshpatel approved these changes Apr 4, 2024
View reviewed changes
Copy link
Contributor

@tirthasheshpatel tirthasheshpatel left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks good, thanks for the fix! I haven't tested with Gemma but LLaMA and Mistral work.

@mattdangerw
Copy link
Member Author

Looks good, thanks for the fix! I haven't tested with Gemma but LLaMA and Mistral work.

Thanks! I am testing with Gemma now.

@mattdangerw
Copy link
Member Author

I think this looks good to go, but to be safe I will probably wait till after cloud next to push our later next week.

@mattdangerw mattdangerw mentioned this pull request Apr 6, 2024
Closed
@mattdangerw mattdangerw marked this pull request as ready for review April 8, 2024 20:17
@mattdangerw mattdangerw merged commit ab649f5 into keras-team:master Apr 10, 2024
12 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@tirthasheshpatel tirthasheshpatel tirthasheshpatel approved these changes

Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.

Why not use low precision matmul for reverse embedding in gemma model
2 participants
@mattdangerw @tirthasheshpatel